After this lesson, you will be able to: Compare Azure Kubernetes Service (AKS) to self-hosted Kubernetes and to EKS/GKE; understand AKS-specific features.
AKS is Azure's managed Kubernetes. Same K8s API as EKS / GKE, with Azure-specific niceties.
Managed control plane (free; you don't pay for the masters). You pay for: worker nodes (VMs), Azure storage / load balancers attached, egress. Supports CNI: Azure CNI (each Pod gets a VNet IP), Kubenet (cheaper but less integrated), Calico, Cilium. Auto-upgrade channels, automatic node patching, autoscaling (Cluster Autoscaler + KEDA for event-driven).
5-minute provision.
# Create resource groupaz group create -n rg-aks -l eastus# Create AKS clusteraz aks create \--resource-group rg-aks \--name myaks \--node-count 2 \--node-vm-size Standard_B2s \--enable-managed-identity \--generate-ssh-keys# Connect kubectlaz aks get-credentials --resource-group rg-aks --name myakskubectl get nodes
Workload Identity: same idea as IRSA on EKS. Pods get short-lived Azure AD tokens via federated identity. Use for Pods accessing Storage / KeyVault / etc. Application Routing Add-on: managed NGINX Ingress + cert-manager out of the box. Container Storage Interface (CSI) drivers: Azure Disk + Files + Blob via CSI. Azure Monitor for Containers: metrics + logs from every Pod into Log Analytics. KQL queries against them. Microsoft Entra integration for kubectl auth: humans authenticate via their corporate identity.
Free control plane (AKS), but Standard SLA tier ($0.10/hr) is recommended for prod. Nodes: VM-priced. A 2x Standard_B2s cluster is ~$60/mo. Load Balancers: ~$25/mo each (Standard tier). Storage: per-disk for PVCs (~$1.50/GB/mo for Premium SSD). Free tier covers Cluster Autoscaler; KEDA is open-source / free.
Same K8s API; same kubectl. Workload portability is high. Pick by org's cloud commitment. If most spending is on Azure, pick AKS. AKS strengths: Entra ID integration, Application Routing Add-on, Container Storage Interface for Files / Blob. AKS weaknesses: ecosystem still slightly behind EKS in vendor integrations; Karpenter equivalent (Node Autoprovisioning) is newer. GKE remains the most polished managed K8s; AKS is catching up fast.
Skipping Workload Identity. Pods then need stored credentials. Modern AKS sets this up via Azure AD trust. Running prod on Free tier control plane. Upgrade to Standard for SLA. Forgetting node pool taints + tolerations for workload isolation. Mixing too many tools (manual kubectl + Helm + IaC). Standardize on one (usually IaC + Helm). No cluster autoscaler. Pods get stuck Pending under load.
Pick the keyless option.
Sign in and purchase access to unlock this lesson.